Tweet
Last year's GeForce GTX 680 rewrote the rule book on the levels of gaming performance we should expect from a high-end single-chip graphics card, seeing off AMD's competitors in terms of power while delivering a remarkable level of quietness and efficiency. Less than 12 months on, the firm has topped its own considerable achievement with the release of the GeForce Titan - a consumer-level edition of its £2,800 Tesla K20 "supercomputer" board. The good news is that the gaming version of this phenomenal technology retails for a fraction of the cost with no appreciable cutbacks in its overall capabilities, but the bad news is that it's still twice the price of the GTX 680. The question is, can the case be made for an £800 graphics card? Nvidia reckons so, its marketing positioning Titan as a luxury product that sits atop its current range: ultimate performance at a stratospheric price tag.Initially, the "Kepler" architecture rolled out in 1536 and 384 CUDA core configurations, aimed at the high-end and mobile/entry-level markets, with mid-range offerings gradually filling in the gap between the two. However, it turns out that the GK104 chip found in the GTX 680 was originally planned to occupy the mid-range space, with another, larger piece of silicon initially slated for the top-end consumer product. When it became apparent that GK104 outpaced AMD's best offerings by quite a margin, it was repositioned as the high-end release, with the original design for the GTX 680 repurposed for the Tesla "supercomputer" line.
Now that larger, more powerful chip - dubbed GK110 - has finally been released into the consumer space, and it's no exaggeration to suggest that in performance terms it's a bit of a monster. Where the GTX 680 features 1536 CUDA cores, the Titan boasts a colossal 2688 - a 75 per cent increase. Onboard GDDR5 RAM gets a threefold boost from the reference design GTX 680's 2GB up to 6GB, while bandwidth is expanded with the move from a 256-bit bus to a meatier 384-bit interface. Transistor count more than doubles from 3.5 billion to 7.1 billion, while ROPs are boosted from 32 to 48, enhancing the card's capabilities in servicing ultra-high resolutions.
Read more…
More...